Learning about Outgroups: Comparing Deep versus Broad Connections

Last registered on March 15, 2024


Trial Information

General Information

Learning about Outgroups: Comparing Deep versus Broad Connections
Initial registration date
March 02, 2024

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
March 15, 2024, 2:40 PM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.



Primary Investigator

Duke University

Other Primary Investigator(s)

PI Affiliation
UC Davis
PI Affiliation
PI Affiliation

Additional Trial Information

On going
Start date
End date
Secondary IDs
Prior work
This trial does not extend or rely on any prior RCTs.

Existing intergroup contact interventions involve sustained interactions with a fixed set of outgroup individuals – such as in fixed sports teams, college dorms etc. A common finding is that such contact changes behavior toward the individuals interacted with, but not help change attitudes towards or dispel stereotypes about the outgroup at large. We will test whether an alternative type of contact causes more generalized effects: briefer contact with a rotating cast of outgroup members. This can inform policy regarding effective ways to integrate individuals in different settings. We will recruit young Hindu and Muslim individuals to work on simple production tasks in West Bengal, India. We will randomize each participant to (1) be paired with an outgroup or ingroup member, and (2) have their partner change every day or stay the same.
External Link(s)

Registration Citation

Nellis, Gareth et al. 2024. "Learning about Outgroups: Comparing Deep versus Broad Connections." AEA RCT Registry. March 15. https://doi.org/10.1257/rct.13126-1.0
Experimental Details


Intervention Start Date
Intervention End Date

Primary Outcomes

Primary Outcomes (end points)
Quantity and quality of production (small and big bags) measured each day of the intervention, controlling for lead-in individual productivity

In a holdout survey with individuals who are not part of the main study, we will measure the behaviors and beliefs of Hindus and Muslims through the following games:
1. (Mis)-reporting the no. of heads in a private coin-toss where each head earns you money (dishonest behavior/ lying)
2.Giving to religious outgroup in a Dictator game (altruism)
3. Movie choice (romantic comedy versus horror) to measure timidness
4. Use of hand sanitizer to measure sanitary practices
5. Lottery game (choice over lottery versus sure payment) to measure risk aversion
6. Self-reported religious versus Indian identity (as well as through choice over bags with religiously distinctive colors versus national flag colors)
7. Self-reported desired number of children (to get at stereotypes around the threat of a growing outgroup population)
8. At the endline, we will incentivize treatment and control participants to correctly predict the behavior of the holdout survey participants to assess how treatment affects generalized belief updating (i.e. updating about people not met) on these dimensions. For some of the questions, we will ask treatment participants to make predictions about their partners at the factory (whereas control participants will also be asked to make predictions about those who worked at the factory for all 9 days) to benchmark effects.
Beyond the outcomes mentioned above, we will also elicit beliefs about the productivity of ingroup and outgroup members after the 3-days of lead in and then again at endline.

Dictator giving to outgroup individual

Self-report whether more attached to being an Indian vs. being a Hindu/Muslim

– Last names of five closest friends (to be coded as Muslim- vs. Hindu-sounding)
– Willingness to work in the future with stranger from outgroup (incentivized)

[NOTE: At the time of submitting this trial registration 14 people had completed the endline survey, but no data had been accessed by any of the research team.]
Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)
– Income earned, hours worked in the last week before endline survey
– Whether factory partners helped get information on or get additional work [TREATMENT INDIVIDUALS ONLY]
– Lent or borrowed money from factory partners [TREATMENT INDIVIDUALS ONLY]

– Daily mood surveys to assess emotions (happiness) during production, rate own + partner’s performance, as well as identity fusion with partners

– Recall names of teammates during the intervention
– Overall happiness during the intervention period across all days
-- Number of bags made as a team

– Thermometer rating for India’s Prime Minister
– Rankings of political systems

– Type of work partner preference
– Willingness to accept outgroup as neighbor (self-reported)
– Contact with partners from the factory and whether time spent on various activities (meeting in person, time spent with each other’s families, meals together, chatted via phone or text message)
– Report phone number of partners from the factory
– Interest in attending a social event
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
We are recruiting Hindus and Muslims aged 18 and above in the city of Barasat (North 24 Parganas) in West Bengal to make paper bags. We have rented a local warehouse for the intervention. Participants must first complete two baseline surveys (one at their home and one at the study site) to be eligible for the study. Afterward, eligible participants will be assigned to work individually on bag production for a 3-day “lead-in” period. Those who complete all 3 lead-in days will then be randomized into the following groups:

(i) A control group who will not receive additional work but will participate in the endline survey
(ii) Deep-Ingroup: Workers will work in pairs for the next 6 days. This group comprises only Hindu workers who will have fixed partners (also Hindus).
(iii) Deep-Outgroup: Workers will work in pairs for the next 6 days. They will have a fixed partner for all 6 days and they will be from the religious outgroup (so all pairs are Hindu-Muslim).
(iv) Broad-Outgroup: Workers will work in pairs for the next 6 days. Their partners will be different each day and they will always be from the religious outgroup (again, all Hindu-Muslim pairs).

Our primary interest is to understand whether different types of exposure (with a fixed outgroup member versus a rotating cast of outgroup members) affect team production, change prejudicial attitudes and behaviors, and help dispel stereotypes. We will also pool across all treatment groups (and compare against the control group) to understand the overall impact of contact in general, and to understand how work affects the formation of social networks and other labor market outcomes.

For our analysis, we will run OLS regressions with randomization strata fixed effects, and a baseline-measured dependent variable when available. We will use robust standard errors when the unit of observation is the individual, and cluster standard errors at the pair-level when the outcome is at that level (such as team production). We will also cluster standard errors at the individual-level when there are multiple observations per individual.
Experimental Design Details
Not available
Randomization Method
Randomization will be done wave-by-wave in STATA. The main randomization after the lead-in period to the control group or 6 further days of work (i.e., to the treatment groups (ii), (iii) and (iv) above) will be stratified by religion. Since workers in the broad-outgroup treatment will have rotating partners each day, we will create their schedules using the following algorithm:
1. Consider all possible inter-religion pairs in the broad-outgroup treatment
2. Randomly sort these pairs
3. Randomly draw the number of pairs required to work each day (repeat if an individual appears in more than one team until this is not the case)
4. Repeat 1-3 for the next days with the additional condition that pairs from previous days are not repeated
Randomization Unit
Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters
Approx. 1,000 individuals (750 in the three treatments and 250 in control), across roughly 10 production waves.
This target sample size is funding permitting, and one or two additional waves may be possible if additional grants come in. Given this, the expected sample range is 800-1,200.
Sample size: planned number of observations
Approx. 1,000 endline response outcomes (assuming no attrition), and 375 team production outcomes (750 workers always working in teams) for 6 days of the treatment, so in total approximately 2,250 team production outcomes
Sample size (or number of clusters) by treatment arms
Roughly equal proportions of each treatment arm (so N=250 for each of the four treatments). However, the randomization and sampling approach will lead to some deviations from equal shares (e.g., if we over-recruit for one wave, that wave will have a higher share of pure control individuals than 25%).
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

Institutional Review Boards (IRBs)

IRB Name
Duke Campus IRB
IRB Approval Date
IRB Approval Number
IRB Name
IRB Approval Date
IRB Approval Number
IRB Name
UC Davis IRB Administration
IRB Approval Date
IRB Approval Number
IRB Name
IRB Approval Date
IRB Approval Number